Abstract

This paper is a description of our system developed for SemEval-2016 Task 9: Chinese Semantic Dependency Parsing. We have built a transition-based dependency parser with online reordering, which is not limited to a tree structure and can produce 99.7% of the necessary dependencies while maintaining linear algorithm complexity. To improve parsing quality we used additional techniques such as pre- and post-processing of the dependency graph, bootstrapping and a rich feature set with additional semantic features.

Highlights

  • Dependency parsing is one of the core tasks in natural language processing, as it provides useful information for other NLP tasks

  • The core of our system is a transition-based dependency parser with on-line reordering in style of Titov et al, (2009)

  • We have built a transition-based semantic dependency parser with online reordering, bootstrapping, additional semantic features and graph pre- and post-processing that achieved the best results in SemEval-2016 Task 9

Read more

Summary

Introduction

Dependency parsing is one of the core tasks in natural language processing, as it provides useful information for other NLP tasks. On the other hand, deals with acyclic graphs, where words may have multiple incoming dependencies. It significantly complicates the task and requires development of the new algorithms or special adoption of the old ones. The main advantage of the transition-based parsers is that they are in general faster than graph-based ones because of the linear complexity of the algorithm (Nivre, 2003). Attardi (2006) introduced additional actions that add dependencies between the roots of non-adjacent subtrees. Both techniques maintain linear algorithm complexity at the expense of incomplete coverage of all possible dependency trees. Complete coverage of all non-projective trees is achieved by Nivre (2009) with a technique called on-line reordering but it increases the worst-case complexity from linear to quadratic

System Description
Parsing Algorithm
Bootstrapping
Semantic Features
Dependencies Pre-processing and Postprocessing
Experiments
Findings
Conclusion
Full Text
Paper version not known

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call